Dispatch from
the offensive side.

Exploitation is a solved problem, discovery isn’t

Mythos is impressive, but shortly after, smaller models replicated similar results by pointing it to specific attack vectors. Our lab tests confirm even open source models perform as reliably as humans in exploitation (with growing reliability over time). However, the key challenge remains discovery, as context is a limited resource for LLMs, which is especially challenging for black box assessments with unlimited attack vectors.